Surveillance robotics: analyzing scenes by colors analysis and clustering

We describe a surveillance system which allows a mobile robot, operating within a familiar setting, to detect unexpected changes in the environment (i.e., presence or absence of objects or persons, unexpectedly opened/closed doors or windows, etc.). To achieve this, the robot "looks at" the environment through a TV camera; next, it compares what "it sees" in a specific moment with what "it should see" at that same location. In particular, an approach is proposed for images comparison which requires to find color clusters in the color histograms corresponding to the images to be compared: by analyzing the color clusters in the two images, the systems detects similarities or differences between them and consequently deduces if something has changed in the scene. Since the techniques proposed in literature usually rely on the geometrical characteristics of the objects to be detected (besides cromination), "ad hoc" algorithms have been implemented for color clusters comparison; the proposed techniques can be used alone or in cooperation with existing pattern recognition methods in order to increase the performance of the whole system.